OPTIMIZATION

gradient descent update

updates parameters along the local steepest descent direction of the objective function.

xk+1=xk−ηk∇f(xk)x_{k+1}=x_k-\eta_k\nabla f(x_k)

symbols, variables and units

x: parameter vector; eta: step size; ∇f: gradient; the unit must make the product consistent with x.

applicable conditions and boundaries

Smoothness and step size affect convergence; non-convex problems do not guarantee global optimality.

formula source code

The following is a copyable LaTeX expression.

x_{k+1}=x_k-\eta_k\nabla f(x_k)

Reference and Extended Learning

Stanford · Convex Optimization, Boyd & Vandenberghe

is organized according to model definition and assumptions. Please check actual conditions and original literature before engineering, research and clinical use.

gradient descentoptimization

How can this knowledge be incorporated into high-end products?

Relevant scientific figures and methodological contributions

Same subject formula

Go to Free Science Tool Library ↗